JMIR Publications

ABSTRACT

Background:

Artificial intelligence (AI) systems are increasingly deployed in digital mental health platforms for early detection of suicide risk and severe psychological distress. Current “responsible AI” approaches often prioritise precision and minimising false positives through human-in-the-loop (HITL) verification. While this can reduce operational strain and perceived liability, it delays interventions in time-critical crises, potentially increasing risk. This trade-off, where greater procedural safety paradoxically increases danger, is termed the Safety Paradox.

Objective:

To introduce the Rapid Intervention Safety Enhancement (RISE) protocol, a framework designed to reduce mean time to intervention (MTI) while maintaining safeguards, and to outline a proposed methodology for its evaluation.

Methods:

The RISE Protocol was developed through iterative design workshops, expert consultations, and review of mental health AI safety literature. It comprises four stages: Rapid Detection, Immediate Triage, Staged Intervention, and Evidence Logging. Each stage includes defined operational targets, intervention pathways, and accountability measures. Key operational metrics are proposed to evaluate system performance.

Results:

As the RISE Protocol has not yet undergone empirical trials, this paper presents it as a conceptual model for future evaluation. An illustrative use case and a comparative analysis against current industry approaches suggest that RISE could enable faster interventions without increasing liability risk, by automating detection and triage to reduce delays from human verification bottlenecks.

Conclusions:

The RISE Protocol reframes mental health AI safety as a function of responsiveness rather than precision alone. By establishing operational standards for mean time to intervention (MTI), cultural adaptation, and accountable automation, it aims to shift the industry toward proactive, life-saving interventions. Future research should focus on empirical validation of the framework and its impact on user outcomes.